Zhang Huibing, Li Tong, Yin Lihua, Liu Dingke, Zhou Ya, Zhang Jingwei, Pan Fang
Guangxi Key Laboratory of Trusted Software, Guilin University of Electronic Technology, Guilin 541004, China.
Cyberspace Institute of Advanced Technology, Guangzhou University, Guangzhou 510006, China.
Sensors (Basel). 2019 Apr 4;19(7):1623. doi: 10.3390/s19071623.
The fusion of multi-source sensor data is an effective method for improving the accuracy of vehicle navigation. The generalization abilities of neural-network-based inertial devices and GPS integrated navigation systems weaken as the nonlinearity in the system increases, resulting in decreased positioning accuracy. Therefore, a KF-GDBT-PSO (Kalman Filter-Gradient Boosting Decision Tree-Particle Swarm Optimization, KGP) data fusion method was proposed in this work. This method establishes an Inertial Navigation System (INS) error compensation model by integrating Kalman Filter (KF) and Gradient Boosting Decision Tree (GBDT). To improve the prediction accuracy of the GBDT, we optimized the learning algorithm and the fitness parameter using Particle Swarm Optimization (PSO). When the GPS signal was stable, the KGP method was used to solve the nonlinearity issue between the vehicle feature and positioning data. When the GPS signal was unstable, the training model was used to correct the positioning error for the INS, thereby improving the positioning accuracy and continuity. The experimental results show that our method increased the positioning accuracy by 28.20-59.89% compared with the multi-layer perceptual neural network and random forest regression.
多源传感器数据融合是提高车辆导航精度的有效方法。随着系统中非线性程度的增加，基于神经网络的惯性器件与GPS组合导航系统的泛化能力会减弱，导致定位精度下降。因此，本文提出了一种KF-GDBT-PSO（卡尔曼滤波器-梯度提升决策树-粒子群优化，KGP）数据融合方法。该方法通过融合卡尔曼滤波器（KF）和梯度提升决策树（GBDT）建立了惯性导航系统（INS）误差补偿模型。为了提高GBDT的预测精度，我们使用粒子群优化（PSO）对学习算法和适应度参数进行了优化。当GPS信号稳定时，采用KGP方法解决车辆特征与定位数据之间的非线性问题。当GPS信号不稳定时，利用训练模型对INS的定位误差进行校正，从而提高定位精度和连续性。实验结果表明，与多层感知器神经网络和随机森林回归相比，我们的方法将定位精度提高了28.20%-59.89%。
